Returns to Sender? Ballot Reform, Partisan Institutions, and the Electoral Consequences of American State-Building

Transcription

1 Returns to Sender? Ballot Reform, Partisan Institutions, and the Electoral Consequences of American State-Building Jon C. Rogowski Harvard University Chris Gibson Washington University in St. Louis July 20, 2016 Abstract The introduction of the Australian ballot is widely credited with ushering in a new era of electoral accountability in American politics, yet existing research provides little direct evidence of its effect on the electoral connection. In this paper, we examine how the Australian ballot affected the relationship between members of Congress and their constituents by studying the relationship between U.S. House elections and the allocation of post offices from 1876 to We find strong evidence that the Australian ballot significantly increased the electoral returns to the allocation of federal post offices. Prior to the introduction of the Australian ballot, we find no evidence that incumbent party vote shares in the U.S. House were responsive to changes in the distribution of post offices. With the introduction of the Australian ballot, however, vote shares were strongly responsive to the provision of post offices. We also find that the effect of the Australian ballot was considerably greater in states with weak party organizations, which suggests that parties with strong electoral organizations could better insulate candidates and officeholders from voter scrutiny when ballot reforms were implemented. Our results have important implications for understanding the conditions under which electoral institutions affect democratic accountability. We thank Alex Bluestone, Michael Byrne, Madeline Josel, Sophie Schuit, and Joe Sutherland for research assistance, and the Office of Undergraduate Research at Washington University in St. Louis for funding. We also thank Cameron Blevins, Cathy Cohen, Justin Fox, Will Howell, Richard John, Sam Kernell, Doug Kriner, John Patty, and Andrew Reeves for helpful discussion and comments. Corresponding author; Assistant Professor, Department of Government, 1737 Cambridge Street, Cambridge, MA 02138; rogowski@fas.harvard.edu. Department of Economics, Campus Box 1208, One Brookings Drive, St. Louis, MO 63130; chris.gibson@ wustl.edu.

2 Representative government turns on the capacity of citizens to sanction and reward elected officials for their performance in office. In designing the American system of government, the Founders specifically designed the executive and legislative branches with an eye toward how the procedures for selecting the president and members of Congress, respectively, would affect those officeholders incentives. For instance, Madison wrote in Federalist 52 that frequent elections are unquestionably the only policy by which members of the House of Representatives would have an immediate dependence on, and an intimate sympathy with, the people. Madison s intuition is reflected in contemporary political science understandings of legislative behavior which largely implicate the incentives provided by the re-election imperative (e.g., Mayhew 1974). Although a sizable literature in political science demonstrates ways in which election outcomes are sensitive to legislative behavior (e.g., Canes-Wrone, Brady and Cogan 2002; Carson et al. 2010), the agency relationship between citizens and legislators was weaker for the country s first century. Nineteenth-century legislators are often believed to have privileged their party s priorities over their constituents interests. As the nineteenth century came to an end, Progressives identified ballot reform as among their top priorities, as the party-printed ballot contributed to corruption and limited voters ability to distinguish candidates from their parties (Engstrom and Kernell 2014; Katz and Sala 1996). In his history of ballot reform, Evans (1917, 15) wrote that party-printed ballots failed to supply any adequate method of acquainting the public with the names of the men they would be called upon the vote for sufficiently in advance of the election to secure an examination into their qualifications. Nominations of corrupt or inefficient men were many times made too late for a public exposure, so the electorate went to the polls relying only on the honesty of their party, which many times betrayed them. According to this account, the party-printed ballot system contributed to the public s lack of knowledge about their party s political candidates and undermined constituents ability to create electoral incentives for legislative behavior. In this paper, we examine how the introduction of the Australian ballot affected the electoral 1

3 connection between members of Congress and their constituents in the late nineteenth century. Building upon studies of the electoral consequences of high-salience nineteenth-century roll call votes (Bianco, Spence and Wilkerson 1996; Carson and Engstrom 2005), we study the connection between U.S. House elections and the allocation of post offices over a twenty-year period from 1876 to Post offices were a key indicator of constituency service and were quintessential distributive goods during a period in which federal policymaking [was] usually of a distributive nature (McCormick 1986, 206). We argue that the adoption of the Australian ballot strengthened the electoral returns to post office allocation by enabling voters to distinguish officeholders on the basis of their performance rather than their party and creating incentives for credit-claiming among legislators. Our study provides among the first systematic examinations of the electoral connection in the nineteenth century and contributes to an active research program on the electoral consequences of distributive politics, virtually all of which focuses on the modern era (Kriner and Reeves 2012; Levitt and Snyder 1997; Stein and Bickers 1994). We find strong evidence that the Australian ballot significantly and substantially increased the electoral returns to the allocation of federal post offices. Prior to the introduction of the Australian ballot, we find no evidence that incumbent party vote shares in the U.S. House were responsive to changes in the distribution of post offices. With the introduction of the Australian ballot, however, vote shares were strongly responsive to the provision of post offices. Our results are robust across a range of model specifications and when accounting for differences in the design of the Australian ballot. We also find that the effect of the Australian ballot was considerably greater in states with weak party organizations, which suggests that parties with strong electoral organizations could better insulate candidates and officeholders from voter scrutiny when ballot reforms were implemented. Our results provide evidence that while some of the conditions for electoral accountability may have been in place in the late nineteenth century, the electoral connection was strengthened with the introduction of the Australian ballot. The results further suggest that the distribution of federal resources such as post offices may have been an important 2

4 strategy by which legislators cultivated personal votes with their constituents. The Electoral Connection and the Nineteenth-Century Congress To what degree do congressional election outcomes reflect legislators performance in office? Virtually all studies on electoral accountability in the U.S. Congress focus on the modern era and report evidence that legislative behavior affects subsequent electoral outcomes. Voters punish legislators for compiling ideologically extreme roll call voting records (e.g., Canes-Wrone, Brady and Cogan 2002), voting too frequently with their congressional copartisans (Carson et al. 2010), voting opposite constituent preferences on a particular issue (Canes-Wrone, Minozzi and Reveley 2011), and supporting the legislative agenda of an unpopular president (Brady et al. 1996). Voters also appear to respond to officeholders behavior in judicial (Gordon and Huber 2007), gubernatorial (Alt, Bueno de Mesquita and Rose 2011), and local school board elections (Berry and Howell 2007). Electoral incentives provide a powerful mechanism by which voters influence elected officials. The core logic, articulated by Madison in Federalist 52, holds that legislators respond to constituency pressures when they perceive an electoral penalty for failing to do so. Carson and Jenkins (2011) identify several features of the political environment that are necessary for this electoral mechanism to exist. Citizens must observe their representatives behavior in office and cast votes based on this information. Legislators must also desire reelection so they have incentives to behave in ways that ensures they remain in office. The combination of these factors creates the conditions under which electoral accountability can occur. The nineteenth-century Congress and the larger political environment that surrounded it differed considerably from the contemporary context. Many voters had limited contact with their member of Congress (Carson and Jenkins 2011), and the prevailing electoral institutions made it difficult to distinguish individual candidates from their party. Election ballots were created 3

5 and separately distributed by the parties themselves. These party-printed ballots listed all of the party s candidates, such that a vote for any one particular candidate was essentially a vote for all of the party s candidates. Furthermore, ballots were deposited publicly, so the general public, including party officials, could see an individual voter s choices. This balloting system discouraged split-ticket voting (Rusk 1970) and thus complicated voters efforts to reward or sanction particular officeholders based on their performance. Thus, down-ballot candidates, such as those running for the House of Representatives, had little incentive to distinguish themselves from other candidates as partisanship was likely the primary decision criteria for most voters (Katz and Sala 1996). Ballot Reform and Electoral Accountability Progressives identified ballot reform as a key way of addressing the stranglehold they perceived parties to have over American politics in the late nineteenth century. In an early publication, political observer Henry George (1883, 211) lamented that democratic government is with us becoming a failure, in large part because parties only operate as to the heads of tickets. It is only as to a few of their nominees that party managers need take public opinion into account; the fortunes of the rest, save in very rare instances, depend upon the fate of the head of the ticket (207). According to Evans (1917, 21), party corruption and inattentiveness had thoroughly frustrated constituents and thus created a favorable attitude on the part of a majority of voters toward reform. George (1883, 208) argued that the Australian ballot would be the greatest single reform and would very much lessen the importance of party nominations and party machinery in the selection of elected representatives. The Australian ballot was first adopted statewide in Massachusetts in 1888 and, in a tribute to the success of Progressive reformers and anti-party sentiment, quickly spread around the country. Less than a decade later, 36 states had implemented the Australian ballot. We test the hypothesis that the Australian ballot strengthened the linkages between legislative 4

6 behavior and election outcomes. The introduction of the Australian ballot beginning in the late nineteenth century is widely credited with marking a turning point for electoral accountability in American politics. The Australian ballot was implemented in most states over a roughly twentyyear period beginning in 1888, though several other states did not introduce the Australian ballot until after According to Katz and Sala (1996, 21), The states conversion from party-strip balloting to Australian (secret) ballots in the 1890s profoundly altered the electoral environment faced by incumbent members of Congress. After the implementation of the Australian ballot, voters viewed all parties candidates on a single ballot, which they cast privately. These reforms enabled voters to split their votes across party lines and select candidates on the basis of their personal characteristics (Katz and Sala 1996). Candidates for political office could then distinguish themselves from their opponents (Engstrom and Kernell 2014). 2 A model by Primo and Snyder (2010) further suggests that the decrease in the parties electoral importance may have both affected the incentives for legislators to secure distributive goods in search of a personal vote and the electoral rewards those goods conferred. As Reynolds and McCormick (1986, 836) succinctly summarizes the accepted wisdom, The Australian ballot thus weakened the party machine and encouraged electoral independence. But despite the importance attributed to the Australian ballot for shaping legislators electoral incentives, scholars have assembled only indirect evidence in support of its effects on electoral accountability. Upon the introduction of the Australian ballot, members of Congress saw benefits to increased tenure on committees (Katz and Sala 1996) and valued some committees more than others (Wittrock et al. 2008). Consequently, the provision of pork barrel spending on rivers and harbors projects increased dramatically from 1888 to 1901 (Wittrock et al. 2008), as legislators saw incentives to direct distributive resources to their districts. Members of Congress also exhibited greater independence in their roll call voting behavior, as party discipline (Carson and Sievert 1 This latter group of states included New Jersey, North Carolina, South Carolina, Georgia, and New Mexico. 2 Ansolabehere, Snyder and Stewart (2000) show that the personal vote increased around the turn of the twentieth century, which corresponds with the time period in which the Australian ballot was introduced. 5

7 2015) and party unity (Wittrock et al. 2008) both decreased. Voter behavior also changed, resulting in reduced turnout (Heckelman 1995) and increased ticket-splitting (Rusk 1970). Though these patterns are certainly suggestive of increased rates of electoral accountability, existing literature provides no direct assessment of whether the Australian ballot strengthened the relationship between legislative behavior and electoral fortunes. Distributive Politics and the Electoral Connection We study how the Australian ballot affected the electoral connection through the lens of distributive politics. Mayhew (1974) identified credit-claiming as one of the key behaviors of reelection-seeking legislators. Distributive resources offer legislators precisely this opportunity, and securing them is one such way legislators demonstrate their attentiveness and service to the district. A great deal of research in the modern era focuses on how federal resources are distributed among members of Congress. These accounts variously attribute the ability of wellpositioned legislators, such as key committee members (Weingast and Marshall 1988), members of the majority party (Cox and McCubbins 2005), and a president s copartisans (Berry, Burden and Howell 2010) to secure disproportionate shares of distributive resources for their constituents, while other scholars argue that distributive politics are sustained in Congress through norms of universalism (Weingast 1979). Though the strength of the evidence varies in support of these accounts, these theories all make clear that legislators perceive clear incentives to securing them for their constituents. Americans expressed clear demands for distributive resources in the nineteenth century. According to McCormick (1979, 286), [d]istributive decisions may have been roughly what the American people wanted, but the details of such policies perpetually fueled conflict. Federal resources were a valuable currency for members of Congress, and disagreement over the provision of these benefits inspired much debate, both within Congress and between Congress and execu- 6

8 tive branch agencies. The conflict over the distribution of federal resources implies that members of Congress perceived benefits to securing them, and the provision of federal projects, such as roads, bridges, and post offices, enabled members of Congress to claim credit for their work on behalf of their constituents. Though not directly focused on the electoral consequences of these projects, Wilson (1986, 735) provides substantial anecdotal evidence that constituent demand was an important factor in legislators incentives to secure rivers and harbors projects around the turn of the twentieth century and concludes that legislators were aware that pursuing the political pork barrel could have a salutary effect on prolonging a legislative career. Other accounts from the late nineteenth century provide similar characterizations of legislators electoral incentives. In perhaps an early statement of Mayhew s thesis, Bryce (1995 [1888], 197) observed that [a]n ambitious congressman is therefore forced to think day and night of his re-nomination, and attempts to secure it with grants from the Federal treasury for local purposes. Existing research on either the nineteenth-century Congress or the returns to distributive outlays provides limited evidence of systematic patterns of electoral accountability. While scholars have identified several specific instances in nineteenth-century congressional politics, such as votes on specific roll call votes and in specific sessions of Congress that suggest the presence of an electoral connection (e.g., Bianco, Spence and Wilkerson 1996; Carson and Engstrom 2005; Jenkins and Weidenmier 1999; Jenkins and Nokken 2008), it is less clear whether the findings from these relatively high-profile episodes translate into more generalizable patterns. And while scholars have long been interested in the relationship between distributive goods and election outcomes, evidence in support of an association is inconsistent across data sources, empirical approaches, and levels of government (Alvarez and Saving 1997; Christenson, Kriner and Reeves 2014; Crespin and Finocchiaro 2013; Kriner and Reeves 2012; Lazarus and Reilly 2010; Levitt and Snyder 1997; Stein and Bickers 1994) and all of it focuses on the modern era. More generally, neither body of literature considers how changes to electoral institutions affected the connection between constituents and lawmakers. 7

9 Data and Methods We study how the introduction of the Australian ballot affected the relationship between the allocation of federal post offices and congressional election outcomes in the late nineteenth century. The dependent variable in our analysis is constructed from data on county-level voting patterns in congressional elections. 3 Using this measure, we calculated the share of the two-party vote cast for the incumbent party of the county s House representative. We characterize the provision of federal resources using the county-level distribution of federal post offices from 1876 to These data come from Rogowski (2016) and were collected from the United States Official Postal Guide in even years from 1876 to Across the entire time period, the mean number of post offices per county was just over 20, and the median county had 16. The county-level distribution is positively skewed; while each county had at least one post office, several counties had well over 100. The distribution of post offices, moreover, varied considerably both between and within counties and states. For most of the nineteenth century, the post office was the single most prominent indicator of the American state in local communities (John 1995). Post offices were desired by and distributed to every corner of the country, and were the only state organization that interacted with all citizens (Carpenter 2001, 66). During this time period, the post office expanded dramatically, from approximately 36,000 post office locations in 1876 to more than 69,000 by Accordingly, the post office constituted an increasingly large proportion of federal expenditures, reaching 18 percent of the domestic federal budget by The late nineteenth-century post office had many of the characteristics of distributive goods. Local communities expressed strong demands for federal post offices, and legislators appeared to go to great length to provide them. Members of Congress, the president, and the Postmaster General s office all received thousands of petitions from local communities requesting the estab- 3 These data come from ICPSR study #8611, Electoral Data for Counties in the United States: Presidential and Congressional Races,

10 lishment of a new post office, 4 and newspaper accounts routinely detailed how local residents mobilized to contact the government regarding the establishment of post offices. 5 Owing to the intense public pressure for post offices, Congress jealously guarded its prerogative to designate post offices and post roads (Mashaw 2012, 267), and legislators frequently contacted the Postmaster General s office on behalf of constituent requests for post offices. 6 Congressional debates over post office politics were routine and often characterized by partisan and regional divisions. One such instance led the New York Times to criticize the crazed legislators for whom temper and judgment [had] departed. 7 Another newspaper observed that Post-office fights are among the most disagreeable subjects [House] members encounter in their careers, and speculated that Rep. Albert Willis of Kentucky will have a hard fight for renomination on account of his decision in the Louisville Post-office case. 8 For members of Congress, securing post offices was an important means of providing constituency service. While federal outlays, which are frequently used in studies of distributive politics in the contemporary era, generally require the insertion of a line-item for a relatively minimal amount of funds, legislators had to exert considerable costly effort to secure post office locations for their constituencies. Though Congress delegated to the Postmaster General the authority to establish post offices in the Postal Act of 1825, they could only be established where they were served by existing postal roads, over which Congress had the exclusive authority to designate. In addition, Congress was responsible for post office appropriations, which funded postal operations. In practice, these arrangements split control over the creation of post offices between Congress and the Post Office Department. Formal approval for post office locations was 4 These petitions are housed in the National Archives, Post Office Department Reports of Site Locations, , M As but one example, see Asbury Park s Post Office, October 22, 1894, New York Times, page 3. 6 One observer estimated that members of Congress devoted one-fourth to one-third of their time handling postal affairs Kernell and McDonald (1999, 796). 7 The Crazed Legislators, June 14, 1878, New York Times. 8 T.C. Crawford, Post-office Fights: Contests that will Figure in the Fall Campaign, St. Louis Post-Dispatch, May 25, 1886, page 4. 9

11 given by the First Assistant Postmaster General. 9 Securing postal locations often depended on a legislator s success in lobbying these executive branch officials to establish post offices in the legislator s communities. Post offices thus serve as a good indicator of a legislator s investment in providing constituency service (Ashworth and Bueno de Mesquita 2006). We characterize the year in which states adopted the Australian ballot based on the figures reported in Katz and Sala (1996, 25). Massachusetts was the first state to adopt the new ballot, in 1888, and another 37 states followed suit between 1889 and We created an indicator for whether a county was in a state that had the Australian ballot in place. Using counties as the units of analysis requires us to drop two sets of observations. First, we exclude counties that are split across multiple congressional districts. Fortunately, this results in the exclusion of only 1.3 percent of counties during this time period. 10 Second, we exclude states (such as South Dakota and Washington) with more than one congressional seat and who elected at-large members of the House, because it is difficult to know exactly how to assign a county to a particular representative. These counties represented another 1.6 percent of all county observations during this time period. Figure 1 provides an initial cross-sectional examination of how the relationship between post office allocation and electoral outcomes varied based on the use of the Australian ballot. The left plot shows the relationship for counties in states that did not use the Australian ballot and the right plot shows the relationship for counties in states where the Australian ballot was used. Each plotted point indicates a county. The fitted bivariate regression line and 95 percent confidence intervals are also shown. The correlations clearly differ across the two plots. The relationship between post office allocation and electoral outcomes is negative when the Australian ballot was not used, as the plot on the left shows, but the right plot shows a positive relationship between 9 The Fourth Assistant Postmaster General assumed this responsibility beginning in As expected, the counties that are dropped are disproportionately in urban places. The average urban population was 58 percent in counties split across multiple districts, compared with 10 percent in counties represented by a unique congressional district. 10

12 post office provision and incumbent party vote share. Though these results provide some preliminary evidence in support of how the Australian ballot strengthened the relationship between the provision of post offices and electoral outcomes, the correlations are relatively weak. Moreover, the results from cross-sectional comparisons are subject to many potential confounding variables that complicate identification. Figure 1: Electoral Outcomes and the Provision of Post Offices, Without Australian Ballot With Australian Ballot r = Incumbent party vote share Incumbent party vote share r = Post offices (logged) Post offices (logged) Plots show the number of post offices (logged) along the x-axis and the vote share for the party of the incumbent representative in U.S. House elections. Elections in states without the Australian ballot are shown in the left plot and elections in states with the Australian ballot are shown in the right plot. The solid line is the fitted bivariate regression line and shaded region represents the 95 percent confidence interval. Accordingly, we use our panel data to conduct a within-county analysis of the relationship between post offices and election outcomes. Following related research on the electoral implications of federal spending (e.g., Kriner and Reeves 2012), our dependent variable is the percentage-point change in county-level incumbent party vote share from one election to the next. This construc11

13 tion focuses our attention on how the provision of post offices affected election outcomes relative to the baseline level of support it granted for a party or candidate in the previous election and helps reduce concerns about potential endogeneity. We operationalize the provision of post offices as the percent change from one election to the next. This characterization helps account for the fact that an increase of one post office, for instance, was likely to have quite a different impact in a rural county that previously had only one or two post offices, compared with a county in a more developed place that had already had dozens of post offices. Across the entire time period, the average increase in post offices over a two-year period was 9 percent, with a standard deviation of 25 percent. 11 Using these variable constructions, Figure 2 compares the changes in post office allocation and incumbent party vote shares in Pennsylvania between 1894 and Pennsylvania adopted the Australian ballot in Darker colors indicate counties that received the largest increases in post offices (left map) and incumbent party vote shares (right map). The maps display fairly wide variation across counties both in the change in post office allocation and incumbent party electoral fortunes. The maps also show some evidence of a connection between post office allocation and election outcomes. Some of the counties in the southeastern corner of the state that experienced relatively large increases in the provision of post offices also increased their support for the incumbent party of their House representative. Moreover, several counties along the northern border experienced decreases in post office allocation and provided less support for the incumbent party of their House representative. At the same time, however, the figure also shows that counties along the western edge of the state received large increases in post offices but incumbent parties experienced decreased vote shares. However, the single comparison of post office allocation and election results does not allow us to fully account for idiosyncratic factors 11 A very small percentage of the counties contained in the data experienced percent changes in post offices greater than 1 (100 percent), the vast majority of which had three or fewer post offices in year t 1, and for whom an increase of only a few post offices translated into a large percentage increase. Excluding these observations yields point estimates that are substantially similar to those reported in the text, and does not change any of our substantive inferences. 12

14 that may have affected either or both variables in a particular county or during a particular year. Figure 2: Post Office Allocation and Election Outcomes, Change in post office allocation Change in incumbent party vote share Decreased 5+% Decreased <5% No change Increased <3% Increased 3 5% Increased 5+% 10+% loss 5 10% loss <5% loss <5% gain 5 10% gain 10+% gain The map on the left shows the change in the number of post offices in Pennsylvania between 1894 and The map on the left shows the change in incumbent party vote share in House elections between 1894 and Darker shades indicate larger increases. Empirical Strategy We use our panel data to model the change in county-level incumbent party vote shares (Y i t ) using linear regressions that take the general form: Y i t = β 0 + β 1 Post offices i t + β 2 Australian ballot i t + β 3 (Post offices i t Australian ballot i t ) + α i + δ t P i t +X i t Ω + ɛ i t, where i denotes counties and t denotes years. Post offices indicates the percent change in post offices between election year t and t 1 and Australian ballot indicates whether the secret bal- 13

15 lot was used. Our key parameter of interest is β 3, which characterizes the extent to which the implementation of the Australian ballot changed the relationship between post office allocation and electoral outcomes. If the introduction of the Australian ballot increased the electoral returns to post offices, we expect a positive coefficient. We include county fixed effects (α i ) to account for time-invariant county characteristics that may also affect election outcomes but are unobserved or otherwise excluded from the model. We also include year fixed effects (δ t ) and its interaction with an indicator for the incumbent legislator s partisanship (P i t ), which captures election-specific partisan tides. Thus, the coefficients for the independent variables are identified using within-county changes in the values of these variables. We also account for several control variables (X i t ) that may also explain changes in the distribution of post offices and incumbent vote shares. Most obviously, in all our models we account for the percentage change in county population, which could have affected demand for post offices. 12 We also include several measures of the electoral environment. Because the Electoral College may have led presidents to direct more resources to more competitive states, we follow Kriner and Reeves (2015) and include an indicator (Swing) for states in which the previous presidential election was decided by 10 percentage points or less. Similarly, to account for the possibility that legislators in competitive districts may attempt to direct more resources toward their constituents, we include an indicator (Marginal) for whether the last congressional election was decided by 10 percentage points or less. We also include indicators for whether the current incumbent won in an Uncontested or Open seat election, and whether the state had redrawn its district lines since the last election (Redistricted). 13 Finally, β 0 is a constant term, Ω is a vector of regression coefficients for the control variables, and ɛ i t is a random error term, which we cluster on congressional districts within a given redistricting cycle County population data were obtained from ICPSR study #2896, Historical, Demographic, Economic, and Social Data: The United States, Population figures were linearly interpolated for intercensal years. 13 Redistricting data come from Jeffrey B. Lewis, Brandon DeVine, Lincoln Pitcher, and Kenneth C. Martis. (2013.) Digital Boundary Definitions of U.S. Congressional Districts, [Data file and code book]. Retrieved from on November 20, Note that the fixed effects specification prevents us from including time-invariant control variables that may 14

16 Results Our results are shown in Table 1. The data provide considerable evidence that the Australian ballot strengthened the connection between legislative behavior and electoral outcomes by increasing the electoral returns to the allocation of post offices. The model shown in column (1) includes our key independent variables and a control for population. The results show, first, little evidence of an association between the allocation of post offices and election outcomes in the absence of the Australian ballot. The coefficient for Percentage Increase in Post Offices is negative (- 0.64), small in magnitude, and not statistically distinguishable from zero. The coefficient estimate for states with the Australian ballot in place is positive (1.20) and suggests that the Australian ballot may have contributed to the incumbency advantage by enabling legislators to cultivate personal votes (e.g., Ansolabehere, Snyder and Stewart 2000), but it is also not statistically significant at conventional levels. However, the interaction between the allocation of post offices and the use of the Australian ballot is positive (11.13), large in magnitude, and statistically significant at p <.006, and indicates that incumbent party vote shares were substantially more responsive to post office allocation when the Australian ballot was in place. As column (2) shows, we find similar results when estimating a model that includes an expanded set of control variables. The coefficient estimate for β 3 is again positive and statistically significant, and indicates that electoral outcomes were affected by the provision of post offices upon the introduction of the Australian ballot, but not otherwise. The results also show that incumbent party vote shares significantly reduced in contexts where the incumbent had previously won in an uncontested or open seat election. None of the other control variables, however, is statistically significant at conventional levels. be correlated with election outcomes or the distribution of post offices. For instance, previous research highlights regional differences in the use of nineteenth-century practices such as rotation (Kernell 1977) and the importance of proximity of a legislator s district to Washington, D.C. for explaining legislators involvement in constituency service (Fenno 1978). While a random effects specification would allow us to account for these and other county-specific factors, results of Hausman tests reject the use of random effects at p <

17 Table 1: The Australian Ballot and the Electoral Returns to Post Offices Independent Variables (1) (2) Dependent variable: Change in incumbent party vote share Percentage Increase in Post Offices (1.20) (0.97) Australian Ballot (1.84) (1.67) Percentage Increase in Post Offices Australian ballot (4.02) (3.71) Percentage Increase in Population (7.35) (7.13) Swing state 0.91 (1.23) Marginal district 0.22 (0.93) Uncontested election (2.13) Open seat 3.45 (0.83) Redistricted 0.48 (0.97) (Intercept) (2.33) (2.07) Observations 18,751 18,751 County fixed effects Year fixed effects Year-party fixed effects Data are from the 1876 through 1896 congressional elections. Entries are linear regression coefficients and standard errors, clustered on congressional district. The dependent variable is the change in vote share received by the incumbent party in percentage points. County, year, and year by party fixed effects are also included. Though the Australian ballot was widely encouraged by Progressive reformers, we do not find any evidence that our results are driven by states with especially Progressive tendencies. 16

18 First, we estimated models that excluded observations from the first states to adopt the Australian ballot. The politics of these states may have been especially Progressive, which could raise concerns about the endogeneity of the policy change or the possibility that these states also adopted other Progressive reforms that could confound the results. We excluded observations from Indiana, Massachusetts, Minnesota, Missouri, Montana, Rhode Island, and Wisconsin, all of whom adopted the Australian ballot before In these models, the coefficient estimates for the Australian ballot and its interaction with post office allocation are identified using the 32 other states that adopted the Australian ballot between 1890 and Second, because Progressive state-level reforms were generally (though not exclusively) advocated by Democrats, we estimated models that excluded states in which the Democratic party was dominant in state politics, which we proxied using states in which both legislative chambers had Democratic majorities. Both of these supplementary analyses provide strong support for the results shown in Table 1 in which the introduction of the Australian ballot substantially increased the electoral returns to post office allocation. Figure 3 displays how the introduction of the Australian ballot affected the responsiveness of vote shares to post office allocation. Based on the estimates shown in column (2) of Table 1, the figure shows the predicted values of the change in incumbent party vote shares across a range of values for the change in post allocation allocation. The solid line shows the predicted change in incumbent party vote share for states without the Australian ballot and the dashed line shows the predicted change in incumbent party vote share for states that used the Australian ballot. The shaded areas show the 95 percent confidence intervals associated with the predicted values. As the solid line shows, the estimates from 1 imply that post office allocation had virtually no effect on election outcomes in states without the Australian ballot. The line is virtually flat across the entire range of the x-axis. The dashed line, however, shows that election results were quite responsive to provision of post offices in states where the Australian ballot was adopted. Consider two otherwise identical counties where one county experienced no increase in post offices and 17

19 the other experienced an increase of 25 percent (which corresponds roughly to one standard deviation). Without the Australian ballot, the incumbent party would be predicted to gain 0.2 percentage points in vote share (p =.87). With the Australian ballot, however, the incumbent would be predicted to gain 4.5 percentage points in vote share (p =.01). Moreover, because an increase in vote shares for one party is associated with a decrease in shares received by the opposite party, a gain of 4.5 percentage points translates to a vote swing of 9 percentage points large enough to have implications for a large number of elections in this period. The difference in the electoral returns to post offices based on the use of the Australian ballot is statistically significant (p =.02). 18

20 Figure 3: Effect of Post Office Provision on Incumbent Party Vote Shares, With Australian ballot Without Australian ballot Difference in incumbent party vote share (percentage points) Percentage change in post offices Estimates are based on the model shown in column (2) from Table 1. The lines are the predicted values of change in incumbent party vote share across a range of values of the change in the provision of post offices. The shaded regions are the 95 percent confidence intervals. The distribution of values of the change in post office provision in states without the Australian ballot are shown at the bottom and the rug marks at the top show the distribution in states with the Australian ballot. The Australian ballot did not take the same form in all states, however, and previous scholarship distinguishes ballots based on their organization. Some states organized ballots by office, where candidates of all parties were listed together for the office they were seeking, while other states used a party column ballot where all candidates from a given party were listed on the same side of the ballot. The type of Australian ballot used could have had important implications for 19

21 the extent to which the secret ballot increased the returns to post office allocations, as the party column ballot could have provided greater incentives for local parties to mobilize local residents (Engstrom 2012). We estimated the models shown in Table 1 but distinguished states that used office bloc ballots from those using party column ballots, and included their interactions with increases in post office provision. The results are shown in Table 2. Consistent with the findings discussed above, the coefficient estimates for the interactions between each ballot type and the allocation of post offices are positive. While the coefficients for the interactions between post offices and party column ballot are large in magnitude and statistically significant, the interaction between post offices and the office bloc ballot are smaller in magnitude and less precisely estimated. However, we do not find that the coefficients are themselves statistically distinguishable (p =.34 for the results in column 2). On the whole, then, our results suggest that the Australian ballot increased the electoral effects of post offices across its various formats. 20

22 Table 2: Ballot Format and the Electoral Returns to Post Offices Independent Variables (1) (2) Dependent variable: Change in incumbent party vote share Percentage Increase in Post Offices (1.20) (1.07) Office bloc ballot (2.37) (2.02) Party column ballot (1.95) (1.82) Percentage Increase in Post Offices Office bloc ballot (6.46) (6.02) Percentage Increase in Post Offices Party column ballot (5.05) (4.72) Percentage Increase in Population (7.40) (7.18) Swing state 1.04 (1.26) Marginal district 0.21 (0.93) Uncontested election (2.13) Open seat 3.47 (0.84) Redistricted 0.46 (0.99) (Intercept) (2.34) (2.09) Observations 18,751 18,751 County fixed effects Year fixed effects Year-party fixed effects Data are from the 1876 through 1896 congressional elections. Entries are linear regression coefficients and standard errors, clustered on congressional district. The dependent variable is the change in vote share received by the incumbent party in percentage points. County, year, and year by party fixed effects are also included. 21

23 The evidence provided here strongly supports claims about the effect of the Australian ballot in enhancing legislative accountability. After the Australian ballot was introduced, voters could distinguish office-seekers from their party, and the available evidence indicates that voters used this opportunity to reward incumbent legislators for the establishment of additional post offices. These findings are consistent with other work that studies the effect of the Australian ballot on legislators incentives to seek plum committee assignments (Katz and Sala 1996) and to secure other forms of distributive goods such as rivers and harbors provisions (Wittrock et al. 2008). These results also provide direct evidence about the role of the Australian ballot in providing an incumbency advantage, possibly through legislators ability to secure a personal vote, a suggestion raised by Ansolabehere, Snyder and Stewart (2000). Partisan Institutions and the Australian Ballot Though the Australian ballot was quickly adopted by a large majority of states, its effects were likely to vary across them in important ways. Though the Australian ballot was believed to be a way of reducing the influence of party corruption and machines on election outcomes, its effectiveness is doing so likely depended upon the nature of party institutions. Specifically, parties with strong electoral organizations likely had the resources and incentives to engage in other forms of behavior such as mobilization and recruitment that could have counteracted the effects of the Australian ballot. The Australian ballot may have had greater effects in places with more weakly organized parties, however, and where parties had traditionally exercised a weaker hand in attempting to influence election outcomes. The strength of party organizations could have conditioned the degree to which the Australian ballot increased the electoral returns to post office provision. Strong party organizations could have used the party machine to mobilize local voters in support of their party s candidates and recruit candidates on the basis of party loyalty. As Primo and Snyder (2010, 355) write, strong party organizations do much of the work in helping candidates run for office and thus candidates 22

24 have less of an incentive to cultivate personal votes. Candidates in areas with weaker party organizations must go to some length to ensure local voters know them, and thus have greater incentive to publicize their efforts to secure post offices and other goods desired by local residents. Thus, the introduction of the Australian ballot likely created different incentives for candidates and elected officials to cultivate personal votes and disseminate information about their records in office based on the nature of the party organization. If this hypothesis is correct, we would expect the effect of the Australian ballot to be greater in places with stronger party organizations. We test this account by distinguishing states with strong party organizations using the typology presented in Mayhew (1986), who identifies states with traditional party organizations according to a set of five criteria. Traditional party organizations are defined as autonomous, enduring, hierarchical in structure, involved in nominating and electing candidates for many different public offices, and responsive to material incentives. Though Mayhew s study focused on the state of party organizations in mid-twentieth century, he writes (1986, 8) that local patterns of traditional party organizations or its lack go back to the turn of the century and before. Thus, following Primo and Snyder (2010), we characterize states with strong party organizations as the thirteen states that received a 4 or a 5 on Mayhew s five-point typology of party strength. These states include Connecticut, Delaware, Illinois, Indiana, Kentucky, Maryland, Missouri, New Jersey, New York, Ohio, Pennsylvania, Rhode Island, and West Virginia. We estimate the same models as before but with separate regressions for states with weak and strong party organizations. The results are shown below in Table 3. The first two columns show the results for weak party states. In general, the findings closely parallel those shown above. The introduction of the Australian ballot significantly increased the electoral returns to the allocation of post offices, and the results are reasonably consistent whether or not the additional control variables are included. However, as the results in columns (3) and (4) show, these patterns do not hold for states with strong parties. The interaction term is incorrectly signed though imprecisely estimated in both models, and thus provides no evidence that post offices increased vote shares 23

25 when the Australian ballot was introduced in states with strong party organizations. Table 3: Party Organizations, the Australian Ballot, and the Electoral Returns to Post Offices Weak party states Strong party states Independent Variables (1) (2) (3) (4) Dependent variable: Change in incumbent party vote share Percentage Increase in Post Offices (1.32) (1.17) (1.98) (1.92) Australian Ballot (2.75) (2.48) (1.72) (1.74) Percentage Increase in Post Offices Australian ballot (5.80) (5.32) (3.82) (3.62) Percentage Increase in Population (7.23) (6.98) (11.86) (11.57) Swing state (1.68) (1.44) Marginal district (1.64) (0.85) Uncontested election (2.22) (7.71) Open seat (1.27) (0.80) Redistricted (1.58) (1.15) (Intercept) (3.19) (2.71) (2.54) (2.79) Observations 11,702 11,702 7,049 7,049 County fixed effects Year fixed effects Year-party fixed effects Data are from the 1876 through 1896 congressional elections. Entries are linear regression coefficients and standard errors, clustered on congressional district. The dependent variable is the change in vote share received by the incumbent party in percentage points. County, year, and year by party fixed effects are also included. The substantive implications of the results in Table 3 are shown in Figure 4. Using the estimates shown in columns (2) and (4), the figure shows the predicted values of the change in 24